On June 2, Anthropic announced the expansion of Project Glasswing to 150 new organizations in more than 15 countries, with NATO and ENISA as the program’s first multilateral government partners. Glasswing protects critical infrastructure (water, energy, health, communications) from the kind of attacks where AI can be both the weapon and the shield. That same day, Claude Code, the same company’s programming tool, suffered a global outage of almost six hours whose root cause was exactly the kind of failure that most worries experts in agentic AI: subagents that multiplied in an infinite loop until they exhausted the systems.
The paradox has additional layers. The incident occurred the same day Anthropic filed its S-1 with the SEC, the document that precedes its IPO. And there is an absence worth naming: after more than six weeks of operation, no Latin American organization appears on the Glasswing list.
Also today
-
Microsoft launches the MAI family: seven in-house models — MAI-Code-1-Flash is now available in GitHub Copilot and VS Code; MAI-Thinking-1, a reasoning model comparable to Claude Opus 4.6, is in Azure preview for enterprises. None was trained on OpenAI data, which marks the start of Microsoft’s own model stack. Project Polaris will replace GPT-4 in Copilot starting in August.
-
Meta: 10 million weekly conversations between businesses and AI on WhatsApp — The figure was presented at Meta Conversations 2026 (London) and represents 10x growth since January. LatAm led the expansion. Meta is also working on AI clones of business staff for the channel.
-
Google Argentina announces $700,000 for four local AI NGOs — The investment, through the Propel program, seeks to expand access to AI for more than 1,000 civil society organizations. Google projects that AI can generate $43 billion in annual value for the Argentine economy.
-
Six-hour Claude outage on June 2: cause confirmed — Max plan quotas were exhausted in minutes. Anthropic Status confirmed that the problem was Claude Code subagents multiplying in a loop. It happened on the day of the S-1 filing with the SEC.
In the region
Colombia enters the home stretch of its presidential runoff (June 21) with a concrete problem: AI is already active in the campaign, but legal protection will not arrive in time. Law 2502 (which penalizes the use of AI to impersonate identities) is in force, but the specific penalties only take effect in July, ten days after the election. Experts point out that cloned audio is the most urgent threat, above deepfake video, because WhatsApp is the dominant distribution channel and its reach is practically impossible to audit in real time. The contrast with Brazil is sharp: that country approved PL 2338 on May 27 and has five active TSE measures in force for these elections. Against that same backdrop, Meta announces from London that LatAm is one of its main growth markets in conversational AI for businesses, while no country in the region has specific AI regulation for business messaging.
It is also worth mentioning a platform that spent some time under the radar: the Regional Observatory on AI in Education for Latin America and the Caribbean, launched in April by UNESCO, CAF, ECLAC and Tec de Monterrey. It brings together 33 education ministries and yields a figure worth keeping in mind: only 45% of higher education institutions in LatAm have formal guidance on AI, compared with 70% in Europe and North America.
Launches
-
MAI-Code-1-Flash (Microsoft) — Microsoft’s first in-house programming model is now generally available in GitHub Copilot and VS Code, with no OpenAI data in its training. It is the first real point of comparison against other coding tools on the market.
-
MAI-Thinking-1 (Microsoft) — Microsoft’s first in-house reasoning model, with performance comparable to Claude Opus 4.6 according to the company. For now in preview for Azure Enterprise customers, but it signals where the ecosystem is heading.
Threads we’re following
The Glasswing story has a background worth recalling. Anthropic launched the program more than six weeks ago as a cybersecurity initiative for critical infrastructure; today it makes the leap to top-tier multilateral government partners such as NATO and ENISA. What has not changed is the region’s absence: no Latin American organization appears in the program. In a context where Meta is growing in business WhatsApp without regional regulation, Colombia faces elections without an anti-deepfake shield in force, and 55% of LatAm universities have no AI guidance, the question of who defends the region’s digital infrastructure (and with what tools of its own) becomes more urgent with each passing week.
If the same AI that seeks to protect critical systems can collapse for six hours because of an infinite loop of subagents, how solid is the layer of defense that Glasswing offers the countries that join the program?
About this entry. It is generated automatically from public sources, without human review before publication. It may contain errors of interpretation or summary; please check each story against its original source (the links lead there) before citing it or making decisions based on it.
Doble Click is written with Anthropic models.